[X86][SSE] Attempt to pre-truncate arithmetic operations that have already been extended

As discussed on D28219 - it is profitable to combine trunc(binop (s/zext(x), s/zext(y)) to binop(trunc(s/zext(x)), trunc(s/zext(y))) assuming the trunc(ext()) will simplify further

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@292493 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index d27aaa4e68c93c54f244ed1ca2cbf3a4ae5b920e..4e4b2b1a6d3e83895785e409619c690888d9ade6 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -32348,13 +32348,23 @@ static SDValue combineTruncatedArithmetic(SDNode *N, SelectionDAG &DAG,
   EVT VT = N->getValueType(0);
   EVT SrcVT = Src.getValueType();
 
-  auto IsRepeatedOpOrOneUseConstant = [](SDValue Op0, SDValue Op1) {
-    // TODO: Add extra cases where we can truncate both inputs for the
-    // cost of one (or none).
-    // e.g. TRUNC( BINOP( EXT( X ), EXT( Y ) ) ) --> BINOP( X, Y )
+  auto IsRepeatedOpOrFreeTruncation = [VT](SDValue Op0, SDValue Op1) {
+    unsigned TruncSizeInBits = VT.getScalarSizeInBits();
     if (Op0 == Op1)
       return true;
 
+    unsigned Opcode0 = Op0.getOpcode();
+    if ((Opcode0 == ISD::ANY_EXTEND || Opcode0 == ISD::SIGN_EXTEND ||
+         Opcode0 == ISD::ZERO_EXTEND) &&
+        Op0.getOperand(0).getScalarValueSizeInBits() <= TruncSizeInBits)
+      return true;
+
+    unsigned Opcode1 = Op1.getOpcode();
+    if ((Opcode1 == ISD::ANY_EXTEND || Opcode1 == ISD::SIGN_EXTEND ||
+         Opcode1 == ISD::ZERO_EXTEND) &&
+        Op1.getOperand(0).getScalarValueSizeInBits() <= TruncSizeInBits)
+      return true;
+
     SDValue BC0 = peekThroughOneUseBitcasts(Op0);
     SDValue BC1 = peekThroughOneUseBitcasts(Op1);
     return ISD::isBuildVectorOfConstantSDNodes(BC0.getNode()) ||
@@ -32386,7 +32396,7 @@ static SDValue combineTruncatedArithmetic(SDNode *N, SelectionDAG &DAG,
     SDValue Op0 = Src.getOperand(0);
     SDValue Op1 = Src.getOperand(1);
     if (TLI.isOperationLegalOrPromote(Opcode, VT) &&
-        IsRepeatedOpOrOneUseConstant(Op0, Op1))
+        IsRepeatedOpOrFreeTruncation(Op0, Op1))
       return TruncateArithmetic(Op0, Op1);
     break;
   }
@@ -32402,7 +32412,7 @@ static SDValue combineTruncatedArithmetic(SDNode *N, SelectionDAG &DAG,
     SDValue Op0 = Src.getOperand(0);
     SDValue Op1 = Src.getOperand(1);
     if (TLI.isOperationLegal(Opcode, VT) &&
-        IsRepeatedOpOrOneUseConstant(Op0, Op1))
+        IsRepeatedOpOrFreeTruncation(Op0, Op1))
       return TruncateArithmetic(Op0, Op1);
     break;
   }

diff --git a/test/CodeGen/X86/vector-trunc-math.ll b/test/CodeGen/X86/vector-trunc-math.ll
index f7e9fdff81b31a7fe31e5d11403910959b84a02f..d6b2b72fb367e1650dbd2c93bcd8382e18e19a4e 100644 (file)
--- a/test/CodeGen/X86/vector-trunc-math.ll
+++ b/test/CodeGen/X86/vector-trunc-math.ll
@@ -415,54 +415,43 @@ define <16 x i8> @trunc_add_v16i16_v16i8(<16 x i16> %a0, <16 x i16> %a1) nounwin
 define <8 x i16> @trunc_add_v8i32_v8i16_sext_8i8(<16 x i8> %a0, <8 x i32> %a1) {
 ; SSE-LABEL: trunc_add_v8i32_v8i16_sext_8i8:
 ; SSE:       # BB#0:
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3]
-; SSE-NEXT:    psrad $24, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; SSE-NEXT:    pslld $16, %xmm2
+; SSE-NEXT:    psrad $16, %xmm2
+; SSE-NEXT:    pslld $16, %xmm1
+; SSE-NEXT:    psrad $16, %xmm1
+; SSE-NEXT:    packssdw %xmm2, %xmm1
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
-; SSE-NEXT:    psrad $24, %xmm0
-; SSE-NEXT:    paddd %xmm2, %xmm0
-; SSE-NEXT:    paddd %xmm1, %xmm3
-; SSE-NEXT:    pslld $16, %xmm3
-; SSE-NEXT:    psrad $16, %xmm3
-; SSE-NEXT:    pslld $16, %xmm0
-; SSE-NEXT:    psrad $16, %xmm0
-; SSE-NEXT:    packssdw %xmm0, %xmm3
-; SSE-NEXT:    movdqa %xmm3, %xmm0
+; SSE-NEXT:    psraw $8, %xmm0
+; SSE-NEXT:    paddw %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_add_v8i32_v8i16_sext_8i8:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vpmovsxbd %xmm0, %xmm2
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpmovsxbd %xmm0, %xmm0
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vpaddd %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpaddd %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; AVX1-NEXT:    vpmovsxbw %xmm0, %xmm0
+; AVX1-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc_add_v8i32_v8i16_sext_8i8:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vpmovsxbd %xmm0, %ymm0
-; AVX2-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17,20,21,24,25,28,29],zero,zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX2-NEXT:    vpmovsxbw %xmm0, %xmm0
+; AVX2-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero,ymm1[16,17,20,21,24,25,28,29],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_add_v8i32_v8i16_sext_8i8:
 ; AVX512:       # BB#0:
-; AVX512-NEXT:    vpmovsxbd %xmm0, %ymm0
-; AVX512-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
-; AVX512-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512-NEXT:    # kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
+; AVX512-NEXT:    vpmovdw %zmm1, %ymm1
+; AVX512-NEXT:    vpmovsxbw %xmm0, %xmm0
+; AVX512-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
   %1 = shufflevector <16 x i8> %a0, <16 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
   %2 = sext <8 x i8> %1 to <8 x i32>
@@ -2308,61 +2297,41 @@ define <8 x i16> @trunc_mul_v8i32_v8i16_zext_8i8(<16 x i8> %a0, <8 x i32> %a1) {
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pxor %xmm3, %xmm3
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7]
-; SSE-NEXT:    movdqa %xmm0, %xmm4
-; SSE-NEXT:    punpckhwd {{.*#+}} xmm4 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
-; SSE-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; SSE-NEXT:    pmuludq %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE-NEXT:    pmuludq %xmm3, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[1,1,3,3]
-; SSE-NEXT:    pmuludq %xmm2, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
-; SSE-NEXT:    pmuludq %xmm1, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
-; SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
-; SSE-NEXT:    pslld $16, %xmm3
-; SSE-NEXT:    psrad $16, %xmm3
-; SSE-NEXT:    pslld $16, %xmm0
-; SSE-NEXT:    psrad $16, %xmm0
-; SSE-NEXT:    packssdw %xmm3, %xmm0
+; SSE-NEXT:    pslld $16, %xmm2
+; SSE-NEXT:    psrad $16, %xmm2
+; SSE-NEXT:    pslld $16, %xmm1
+; SSE-NEXT:    psrad $16, %xmm1
+; SSE-NEXT:    packssdw %xmm2, %xmm1
+; SSE-NEXT:    pmullw %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: trunc_mul_v8i32_v8i16_zext_8i8:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm2 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; AVX1-NEXT:    vpmulld %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpmulld %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: trunc_mul_v8i32_v8i16_zext_8i8:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
-; AVX2-NEXT:    vpmulld %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17,20,21,24,25,28,29],zero,zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX2-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero,ymm1[16,17,20,21,24,25,28,29],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_mul_v8i32_v8i16_zext_8i8:
 ; AVX512:       # BB#0:
-; AVX512-NEXT:    vpmovzxbd {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
-; AVX512-NEXT:    vpmulld %ymm1, %ymm0, %ymm0
-; AVX512-NEXT:    vpmovdw %zmm0, %ymm0
-; AVX512-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
+; AVX512-NEXT:    # kill: %YMM1<def> %YMM1<kill> %ZMM1<def>
+; AVX512-NEXT:    vpmovdw %zmm1, %ymm1
+; AVX512-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX512-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
   %1 = shufflevector <16 x i8> %a0, <16 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
   %2 = zext <8 x i8> %1 to <8 x i32>
@@ -5131,38 +5100,11 @@ define <4 x i32> @mul_add_const_v4i64_v4i32(<4 x i32> %a0, <4 x i32> %a1) nounwi
 ; SSE-NEXT:    paddd {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX1-LABEL: mul_add_const_v4i64_v4i32:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpaddd {{.*}}(%rip), %xmm0, %xmm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: mul_add_const_v4i64_v4i32:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpaddd {{.*}}(%rip), %xmm0, %xmm0
-; AVX2-NEXT:    retq
-;
-; AVX512F-LABEL: mul_add_const_v4i64_v4i32:
-; AVX512F:       # BB#0:
-; AVX512F-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
-; AVX512F-NEXT:    vpaddd {{.*}}(%rip), %xmm0, %xmm0
-; AVX512F-NEXT:    retq
-;
-; AVX512BW-LABEL: mul_add_const_v4i64_v4i32:
-; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpaddd {{.*}}(%rip), %xmm0, %xmm0
-; AVX512BW-NEXT:    retq
-;
-; AVX512DQ-LABEL: mul_add_const_v4i64_v4i32:
-; AVX512DQ:       # BB#0:
-; AVX512DQ-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX512DQ-NEXT:    vpmovzxdq {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
-; AVX512DQ-NEXT:    vpmullq %zmm1, %zmm0, %zmm0
-; AVX512DQ-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512DQ-NEXT:    vpaddd {{.*}}(%rip), %xmm0, %xmm0
-; AVX512DQ-NEXT:    retq
+; AVX-LABEL: mul_add_const_v4i64_v4i32:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpaddd {{.*}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    retq
   %1 = sext <4 x i32> %a0 to <4 x i64>
   %2 = sext <4 x i32> %a1 to <4 x i64>
   %3 = mul <4 x i64> %1, %2
@@ -5213,38 +5155,11 @@ define <4 x i32> @mul_add_self_v4i64_v4i32(<4 x i32> %a0, <4 x i32> %a1) nounwin
 ; SSE-NEXT:    movdqa %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX1-LABEL: mul_add_self_v4i64_v4i32:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpaddd %xmm0, %xmm0, %xmm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: mul_add_self_v4i64_v4i32:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpaddd %xmm0, %xmm0, %xmm0
-; AVX2-NEXT:    retq
-;
-; AVX512F-LABEL: mul_add_self_v4i64_v4i32:
-; AVX512F:       # BB#0:
-; AVX512F-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
-; AVX512F-NEXT:    vpaddd %xmm0, %xmm0, %xmm0
-; AVX512F-NEXT:    retq
-;
-; AVX512BW-LABEL: mul_add_self_v4i64_v4i32:
-; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpaddd %xmm0, %xmm0, %xmm0
-; AVX512BW-NEXT:    retq
-;
-; AVX512DQ-LABEL: mul_add_self_v4i64_v4i32:
-; AVX512DQ:       # BB#0:
-; AVX512DQ-NEXT:    vpmovsxdq %xmm0, %ymm0
-; AVX512DQ-NEXT:    vpmovsxdq %xmm1, %ymm1
-; AVX512DQ-NEXT:    vpmullq %zmm1, %zmm0, %zmm0
-; AVX512DQ-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512DQ-NEXT:    vpaddd %xmm0, %xmm0, %xmm0
-; AVX512DQ-NEXT:    retq
+; AVX-LABEL: mul_add_self_v4i64_v4i32:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpaddd %xmm0, %xmm0, %xmm0
+; AVX-NEXT:    retq
   %1 = sext <4 x i32> %a0 to <4 x i64>
   %2 = sext <4 x i32> %a1 to <4 x i64>
   %3 = mul <4 x i64> %1, %2
@@ -5256,102 +5171,39 @@ define <4 x i32> @mul_add_self_v4i64_v4i32(<4 x i32> %a0, <4 x i32> %a1) nounwin
 define <4 x i32> @mul_add_multiuse_v4i64_v4i32(<4 x i32> %a0, <4 x i32> %a1) nounwind {
 ; SSE-LABEL: mul_add_multiuse_v4i64_v4i32:
 ; SSE:       # BB#0:
-; SSE-NEXT:    movdqa %xmm0, %xmm2
-; SSE-NEXT:    psrad $31, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
-; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE-NEXT:    movdqa %xmm3, %xmm2
-; SSE-NEXT:    psrad $31, %xmm2
-; SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
-; SSE-NEXT:    movdqa %xmm1, %xmm2
-; SSE-NEXT:    psrad $31, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm1[2,3,0,1]
-; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; SSE-NEXT:    movdqa %xmm4, %xmm2
-; SSE-NEXT:    psrad $31, %xmm2
-; SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
-; SSE-NEXT:    movdqa %xmm3, %xmm2
-; SSE-NEXT:    psrlq $32, %xmm2
-; SSE-NEXT:    pmuludq %xmm4, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,1,1,3]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,1,3,3]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm1[0,1,1,3]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
 ; SSE-NEXT:    movdqa %xmm3, %xmm5
-; SSE-NEXT:    pmuludq %xmm4, %xmm5
-; SSE-NEXT:    psrlq $32, %xmm4
-; SSE-NEXT:    pmuludq %xmm3, %xmm4
-; SSE-NEXT:    paddq %xmm2, %xmm4
-; SSE-NEXT:    psllq $32, %xmm4
-; SSE-NEXT:    movdqa %xmm0, %xmm6
+; SSE-NEXT:    psrlq $32, %xmm5
+; SSE-NEXT:    pmuludq %xmm1, %xmm5
+; SSE-NEXT:    movdqa %xmm1, %xmm6
 ; SSE-NEXT:    psrlq $32, %xmm6
-; SSE-NEXT:    pmuludq %xmm1, %xmm6
-; SSE-NEXT:    movdqa %xmm0, %xmm2
-; SSE-NEXT:    pmuludq %xmm1, %xmm2
+; SSE-NEXT:    pmuludq %xmm3, %xmm6
+; SSE-NEXT:    paddq %xmm5, %xmm6
+; SSE-NEXT:    psllq $32, %xmm6
+; SSE-NEXT:    pmuludq %xmm1, %xmm3
+; SSE-NEXT:    paddq %xmm6, %xmm3
+; SSE-NEXT:    movdqa %xmm2, %xmm1
 ; SSE-NEXT:    psrlq $32, %xmm1
-; SSE-NEXT:    pmuludq %xmm0, %xmm1
-; SSE-NEXT:    paddq %xmm6, %xmm1
-; SSE-NEXT:    psllq $32, %xmm1
-; SSE-NEXT:    paddq %xmm0, %xmm1
-; SSE-NEXT:    paddq %xmm1, %xmm2
-; SSE-NEXT:    paddq %xmm3, %xmm4
-; SSE-NEXT:    paddq %xmm5, %xmm4
-; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm4[0,2]
-; SSE-NEXT:    movaps %xmm2, %xmm0
+; SSE-NEXT:    pmuludq %xmm4, %xmm1
+; SSE-NEXT:    movdqa %xmm4, %xmm5
+; SSE-NEXT:    psrlq $32, %xmm5
+; SSE-NEXT:    pmuludq %xmm2, %xmm5
+; SSE-NEXT:    paddq %xmm1, %xmm5
+; SSE-NEXT:    psllq $32, %xmm5
+; SSE-NEXT:    pmuludq %xmm4, %xmm2
+; SSE-NEXT:    paddq %xmm5, %xmm2
+; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm3[0,2]
+; SSE-NEXT:    paddd %xmm2, %xmm0
 ; SSE-NEXT:    retq
 ;
-; AVX1-LABEL: mul_add_multiuse_v4i64_v4i32:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vpmovsxdq %xmm0, %xmm2
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vpmovsxdq %xmm0, %xmm0
-; AVX1-NEXT:    vpmovsxdq %xmm1, %xmm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX1-NEXT:    vpmovsxdq %xmm1, %xmm1
-; AVX1-NEXT:    vpmuldq %xmm1, %xmm0, %xmm1
-; AVX1-NEXT:    vpmuldq %xmm3, %xmm2, %xmm3
-; AVX1-NEXT:    vpaddq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpaddq %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm2[0,2],xmm0[0,2]
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: mul_add_multiuse_v4i64_v4i32:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vpmovsxdq %xmm0, %ymm0
-; AVX2-NEXT:    vpmovsxdq %xmm1, %ymm1
-; AVX2-NEXT:    vpmuldq %ymm1, %ymm0, %ymm1
-; AVX2-NEXT:    vpaddq %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
-;
-; AVX512F-LABEL: mul_add_multiuse_v4i64_v4i32:
-; AVX512F:       # BB#0:
-; AVX512F-NEXT:    vpmovsxdq %xmm0, %ymm0
-; AVX512F-NEXT:    vpmovsxdq %xmm1, %ymm1
-; AVX512F-NEXT:    vpmuldq %ymm1, %ymm0, %ymm1
-; AVX512F-NEXT:    vpaddq %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512F-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX512F-NEXT:    retq
-;
-; AVX512BW-LABEL: mul_add_multiuse_v4i64_v4i32:
-; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vpmovsxdq %xmm0, %ymm0
-; AVX512BW-NEXT:    vpmovsxdq %xmm1, %ymm1
-; AVX512BW-NEXT:    vpmuldq %ymm1, %ymm0, %ymm1
-; AVX512BW-NEXT:    vpaddq %ymm1, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512BW-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX512BW-NEXT:    retq
-;
-; AVX512DQ-LABEL: mul_add_multiuse_v4i64_v4i32:
-; AVX512DQ:       # BB#0:
-; AVX512DQ-NEXT:    vpmovsxdq %xmm0, %ymm0
-; AVX512DQ-NEXT:    vpmovsxdq %xmm1, %ymm1
-; AVX512DQ-NEXT:    vpmullq %zmm1, %zmm0, %zmm1
-; AVX512DQ-NEXT:    vpaddq %ymm1, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpmovqd %zmm0, %ymm0
-; AVX512DQ-NEXT:    # kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
-; AVX512DQ-NEXT:    retq
+; AVX-LABEL: mul_add_multiuse_v4i64_v4i32:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpmulld %xmm1, %xmm0, %xmm1
+; AVX-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    retq
   %1 = sext <4 x i32> %a0 to <4 x i64>
   %2 = sext <4 x i32> %a1 to <4 x i64>
   %3 = mul <4 x i64> %1, %2